Multi-satellite data fusion for improved field-scale evapotranspiration mapping on Google Earth Engine

Abstract

Accurate field-scale evapotranspiration (ET) data with high spatiotemporal resolution is crucial for characterizing surface energy and water balance dynamics and guiding water resource management. OpenET, implemented on Google Earth Engine (GEE), provides field-scale ET estimates and relies mainly on Landsat data. Although Landsat thermal infrared (TIR) observations are effective for field-scale ET mapping, the ∼8-day revisit interval of the combined Landsat 8/9 constellation provides insufficient temporal sampling for short-term ET dynamics. This study presents a framework to improve the spatiotemporal resolution of ET mapping by integrating TIR observations from ECOSTRESS and VIIRS with Harmonized Landsat-Sentinel (HLS) data on GEE. Land surface temperature (LST) data from Landsat, ECOSTRESS and VIIRS were sharpened to 30-m resolution using the Data Mining Sharpener (DMS) algorithm. These sharpened LST data, along with 30-m Leaf Area Index (LAI) and albedo derived from HLS, were used as inputs to the GEE-based Disaggregated Atmosphere-Land Exchange (DisALEXI) model to produce daily 30-m ET estimates. ET estimates were validated against flux tower observations and compared with baseline Landsat-derived ET at six sites with varying land cover and climatic conditions. Results indicated that incorporating ECOSTRESS and VIIRS generally improved ET estimation accuracy, reducing average MAE (mm/day) by 8.64% (1.12 to 1.02, daily), 14.40% (1.00 to 0.85, weekly), 16.37% (0.82 to 0.69, monthly) relative to Landsat-only baselines. This GEE-based framework establishes a prototype workflow for integrating new satellite data sources into the OpenET modeling framework, supporting sustainable agriculture and water resource management.